The present invention relates to a multi-layer neural network, and more particularly to a self-learning multi-layer neural network and the learning method.
The learning multi-layer neural network performs desired functions in such a manner that a user stores the weight values of the respective synapse in an external memory to perform a learning. In other words, the user stores different weight values in the memory according to input and output functions to be used to let the respective synapse have the corresponding different weight values, thereby performing desired functions. Thus, it was inconvenient that the user stored the weight values possessed by the synapse of the multi-layer neural network in the memory so as to perform the desired functions.